A witness can describe lights above a city while leaving their distance, size, and motion uncertain. An illustration may help a reader understand that description. It can also make the missing details look settled.
That is the central limitation when using Sora or another generative visual tool for historical UFO reports. The result depicts a prompt. It does not provide an additional observation of the event.
Start with traceable accounts
The original project used summaries of the Tehran incident, the USS Nimitz encounter, and the Phoenix Lights. Those topics can be starting points for research, but a summary page should lead to the underlying reports, recordings, and attributed testimony.
Record who described each detail, when the account was made, and whether it was firsthand. Distinguish contemporary records from later recollections. Several summaries repeating one account do not constitute independent corroboration.
Build a small evidence table before generating anything:
| Detail | What to record |
|---|---|
| Shape or light pattern | Exact description and source |
| Position | Observer location, direction, and uncertainty |
| Motion | Reported behavior and any measurement supporting it |
| Environment | Time, weather, terrain, and available records |
| Disagreement | Accounts that describe the scene differently |
The table is a proposed research aid. It should contain actual source references before being used to support an illustration.
Separate reported details from artistic choices
Use the prompt to preserve those categories. If an account describes three lights but does not establish a solid object connecting them, a dark triangular hull would be an added interpretation.
A prompt might ask for “an illustrative night-sky scene showing three lights over desert terrain, with their distance and any connecting structure left unspecified.” That is a hypothetical example. It does not claim consensus about a particular sighting.
Record choices about viewpoint, framing, color, and atmosphere. Words such as “eerie” can influence the viewer’s interpretation without contributing evidence.
Providing reference text to a model also differs from training the model on that material. Keep the description of the process accurate.
Label the generated material
The original post presented the following images as generative illustrations. They are retained with captions that identify that role.

Generated illustration inspired by Phoenix Lights accounts. It is not a photograph or measurement of the event.

Generated illustration associated with descriptions of the Tehran encounter. Depicted appearance and geometry are interpretive.

Generated illustration accompanying witness descriptions. A still image cannot establish reported speed or silence.
Labels should remain visible wherever the images are shared. If the original prompt, model version, and generation settings are unavailable, say so in the project record. Those details were not supplied with this article.
Compare alternatives without treating them as evidence
Several illustrations can show how different interpretations fit an incomplete description. For example, separate lights and lights attached to one object may both be compatible with a brief account.
The model’s ability to render either scene does not favor one explanation. Nor does a visually consistent result establish physical plausibility. Measurements, viewing geometry, and independent records must carry that analysis.
Avoid showing a witness a generated reconstruction as though it verifies their memory. A polished image can introduce details absent from the original account.
Use the method for communication
The appropriate use is to help readers inspect what was reported and what the illustrator supplied. Keep the source account beside the image, explain uncertainty, and retain alternative descriptions when they matter.
A useful reconstruction leaves the audience able to distinguish the observation from the interpretation. If it makes that distinction harder to see, revise the image or its presentation.
Share this article

yankee0one
Decade of experience watching PBS Space Time on YouTube.
Related Articles

3I/ATLAS: An Evidence Log of the 2025 Observations
A historical log of Swift, JWST, SPHEREx, VLT, Keck, and solar-monitoring observations, separating measurements, model-dependent estimates, and unresolved reports.
Read moreWorking with Vera C. Rubin Observatory Data
How Rubin observations become images, catalogs, and alerts, and how to choose an access route and investigate a candidate variable source.
Read more
Mars in 2025: What Has to Work Before a Human Landing
A 2025 perspective on the demonstrations, infrastructure, and mission decisions needed to support people on Mars.
Read more